Drop v0.3 support

[TotalVerb]

I have intentionally avoided touching the tests in this PR, so that it is more certain that nothing has regressed. Tests that are testing functionality in Base can be removed in a later PR.

Closes #277 by addressing it directly.

Also closes the following issues on account of the issues becoming obsolete, as they request a feature or bugfix in Base as of v0.4.0 release:

• closes at-compat macro does not work with semi-colon in "split" #31
• closes Add support for redesign of Triangular #46
• closes Pair() for 0.3? #61
• closes size with multiple dim arguments #77
• closes rem() for integer types #87
• closes Cint not working in 0.3.7 #89
• closes Nullable with subtype #94
• closes @compat for []? #102
• closes lufact pivot argument  #112

Also closes the following PRs on account of the added functionality being present in Base as of the v0.4.0 release:

• closes RFC: Add new eltype definitions for v0.3 compatibility #43
• closes Add size multiple dimensions as tuple #78
• closes add karg remove_destination to mv() #91
• closes Add @fastmath as a noop. #92
• closes Add support for set_zero_subnormals() #120

[tkelman]

I don't see why it's useful to delete all the code.

[andreasnoack]

Less code to keep track of and lower risk of clashes between old and new definitions. Why would like to keep it if we drop support for 0.3?

[stevengj]

I think the argument is that it is will be harder and harder to maintain 0.3 compatibility going forward as we add new 0.6 features. And once we drop 0.3 support for new features, we might as well delete the corresponding code for old features (since 0.3 users can still use the older Compat versions).

[tkelman]

I disagree that deleting the code is worthwhile. It may be necessary to backport something to an 0.3-supporting branch, in which case it's better to have the code diverge less.

[TotalVerb]

The code deletion is not urgent and can be discussed at a later time. I'll rebase it out of this PR.

[TotalVerb]

All the code deletions have been reverted now.